Event detection and aspects in twitter: A bow approach

Abstract

Tweets carry much information on the context people are tweeting. Finding the context of tweets or finding the event the tweets talk of is a hot research problem. Among several techniques like statistical, graph-based, machine learning, NLP based and many such techniques, bag-of-words technique is simple and elegant. This paper reports an event detection technique using clustering of bag-of-words of a given set of tweets. The method proposed follows three phase incremental clustering applying Jaccard similarity and Simpson similarity coefficients at different phases. Further, our method is capable of detecting different aspects of an event using a heuristic called EAAS (Event And Aspects Selection) based on tweeter participation, cluster quality and word commonality with the detected event. As a case study, we have used publicly available tweets, collected from Twitter streaming API with a keyword-based strategy. The obtained event detection result is presented and aspects of an event are evaluated in terms of precision and recall against human annotators. With concluding remark the paper presents its findings the possibility of enhancement in event detection as well as aspect finding capability.